Suggestion on reducing short kernel hangs from my btrfs filesystems: bcache?

I have a server which runs zoneminder (video recording which is CPU and disk IO intensive) while also doing a bunch of I/O over serial ports.

I have a a dual core
Intel(R) Core(TM) i3-2100T CPU @ 2.50GHz
(4 virtual CPUs in /proc/cpuinfo)

It's pretty clear that when zoneminder is doing more work, my programs that talk to serial ports start failing due to delays on the kernel side and desynchronization, causing serial port protocol errors (I'm using USB serial adapters, and use 12 of them).
I'm pretty sure it's because of delays in the kernel more than user space, but can't prove that easily.

I have a preempt kernel, kernel 3.16.3:
CONFIG_TREE_PREEMPT_RCU=y
CONFIG_PREEMPT_RCU=y
CONFIG_PREEMPT_NOTIFIERS=y
# CONFIG_PREEMPT_NONE is not set
# CONFIG_PREEMPT_VOLUNTARY is not set
CONFIG_PREEMPT=y
CONFIG_PREEMPT_COUNT=y
CONFIG_DEBUG_PREEMPT=y

btrfs (not counting times where I resync the software raid5 underneath or run a btrfs scrub).

I may try to see if VOLPREMPT might work better, but I'm thinking putting an SSD in front of that mdadm RAID5 array will help by relieving the IO load and hopefully giving more time for the CPU to handle serial port requests.
I'm actually not sure if my issue is btrfs interrupting serial port connections due to PREEMPT, or if serial port connections aren't being serviced quickly enough because the kernel is busy with btrfs and PREMPT hasn't kicked in yet.

that, I was curious if there are other or better ways to use an SSD to make btrfs less impacting on my server?

Thanks,
Marc

"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
Microsoft is to operating systems ....
                                      .... what McDonalds is to gourmet cooking
Home page: http://marc.merlins.org/                         | PGP 1024R/763BE901