Kernel panic (reiserfs), some hangs - bad hardware?.

I have following problems:

1) (looks like )Applications such as oggenc or flac, who mmap() output files and write into mapped memory area, produce corrupted files on my system. ~1 error on 10-100M, resulting file just some bytes different from what program wrote. Second attempts always removes error (while may cause new one with same proportion)
More errors on intensive IO/bad interface cable, less errors on no other IO/good cable. Tried 2.6.7 vanilla kernel.
Am i crazy?
2) With bad interface cable, my system used to just hang. With better?/40pin cable, it do this very rare.
Memtest86 and burncpu do not produce errors, and windows works fine on same box. Both 2.4.x and 2.6.7 kernels.
3) Main. I have got a reiserfs partition with some errors (due to 2)), and kernel panics on it when i do 'find /'.
with these errors:
==
double fault, gdt at c03d2000 [255 bytes]
double fault, tss at c030e80
eip = c0115116
esp = cd0e98b0768
eax = 00000a96 ebx = cf2a547c ecx = 0f0001e3 edx = 0f000000
esi = 00000000 edi = c0115030
*OR*
EFLAGS: 0010007 (2.6.7)
EIP is at scheduler_tick+0x108/0x400
eax: 00000000 ebx: 00000001 ecx: 0000023d edx: 00000000
esi: ca0a9750 edi: c0343520 ebp: ca0a7d4c esp: ca0a7d30
ds: 007b es: 007b ss: 0068
Process (pid: -1048345824, threadinfo = ca0a6000, task=ca0a9750)

Stack: c188a23c 00145ad7 00000000 3d108500 00000000 00000001 00000000
      ca0a7dd4 c0121606 00000000 00000001 00000001 00000000 ca0a6000
      ca0a7dd4 c0121834 00000000 ca0a6000 c010a2ce ca0a7dd4 c02d51c4
      20000001 00000000 c010632a

Trace:
c0121606 update_process_times+0x46/0x60
c0121834 do_timer+0x34/0xf0
c010a2ce timer_interrupt+0x4e/0x120
c010632a handle_IRQ_event+0x3a/0x70
c01066c1 do_IRQ+0x91/0x130
c0124a48 common_interrupt+0x18/0x20
c01ba80a _mmx_memcpy+0x8a/0x170
d08c8630 reiserfs_readdir+0x4c0/0x560 [reiserfs]
c0127052 in_group_p+0x42/0x80
d08e5fa9 __reiserfs_permission+0x169/0x260 [reiserfs]

Code: 0f ba 68 08 03 83 c4 10 56 5e 5f c9 c3 b8 00 e0 ff ff 21 e0

Panic: Fatal exception in interrupt
==
Both 2.4.x and 2.6.7. Running fsck.reiserfs --rebuild-trees cured problem, but should it behave like this anyway?
I have this kernel and this filesystem image, and will assist if someone will go fix this. All numbers here was transferred by screen-paper-screen, so are not guaranted to be correct.
4) I have sacred cow files on my XFS system (again thanx to 2)).
They exist, but can not be stat()ed and deleted.
like this:
==
(1)/usr/share/locale/gr% LANG=C ls
ls: ESSAGES: No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: d·I"Ps?A0V+ApS+A`3U^?: No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
ls: A"pT--I^Pd·I"?a*RE^PZ+: Invalid argument
ls: : No such file or directory
ls: : No such file or directory
ls: : No such file or directory
==
fsck.xfs looks pretty like void main() {};, what should i do?

P.S. Athlon 1333, Via KT266 (Elitegroup motherboard), Seagate Barracuda (don't actually remember model, ST2x0036A?). Kernel 2.6.7.
Checked with memtest86 (3 passes), cpuburn (25 mins), SeaTools (controller, simple, and full tests) - all OK.

P.P.S. Please CC responces to me.

Perhaps a stack overflow? Change THREAD_SIZE in
include/asm-i386/thread_info.h to 16384.

I do not understand, what do not work - drive, cables, or motherboard. But is this a reason for handing - anyway?

SeaTools do not find any errors, so do windows.
Only linux cause errors on this box.

Did objdump -d kernel/sched.o > kernel/sched.objdump

Did this
So:
==
00000940 <scheduler_tick>:
...

     a28:       99                      cltd
     a29:       03 05 10 00 00 00       add    0x10,%eax
     a2f:       13 15 14 00 00 00       adc    0x14,%edx
     a35:       89 47 10                mov    %eax,0x10(%edi)
     a38:       89 57 14                mov    %edx,0x14(%edi)
     a3b:       a1 08 04 00 00          mov    0x408,%eax
     a40:       39 46 28                cmp    %eax,0x28(%esi)
     a43:       74 10                   je     a55 <scheduler_tick+0x115>
     a45:       8b 46 04                mov    0x4(%esi),%eax
     a48:       0f ba 68 08 03          btsl   $0x3,0x8(%eax)
     a4d:       83 c4 10                add    $0x10,%esp
     a50:       5b                      pop    %ebx
     a51:       5e                      pop    %esi
     a52:       5f                      pop    %edi
     a53:       c9                      leave

...
==
0x940+0x108 is 0xa48, right?
Which is:
==

        cltd
        addl    per_cpu__kstat+16, %eax
        adcl    per_cpu__kstat+20, %edx
        movl    %eax, 16(%edi)
        movl    %edx, 20(%edi)
        movl    per_cpu__runqueues+40, %eax
        cmpl    %eax, 40(%esi)
        je      .L395
        movl    4(%esi), %eax
#APP
        btsl $3,8(%eax)
.L391:
#NO_APP
.L367:
        addl    $16, %esp
        popl    %ebx
        popl    %esi
        popl    %edi

leave
==
in .s-file.

I do not konw how to find relevant piece in C code, i don't know good both i386 assembler and debugging to do so.

But i am almost sure that problem is in reiserfs code, not here, 'cause it happened only with broken reiserfs partition when stat()ing files on it.

Ok, where can i find them?