I was able to reproduce the problem on my 2nd Gen. Unfortunately, the problem tends to go away during intensive testing, and so I can't be 100% sure it was caused by r17097.
I'm quite sure it is though, as I found a bug in the SVN assembler code for PP5002: It also enters the loop and writes a pair of samples if there's no or just one free FIFO slot. I prepared a fix, put a test build on my 2nd Gen, and prepared a test build for alleyoopster (3rd Gen). Attached the patch if someone else wants to test.
Will use the test build tomorrow during the day, and if it really fixes the problem, commit tomorrow night.