I attached a digital logic analyzer @ 24MHz (i.e. sampling rate of 41.666ns), transferred 0xFF 0x00 0x00, here were the results.
T0H: 0.500µs. Extreme tolerance of spec (350ns ± 150ns)
T0L: 0.937µs. Inconclusive whether this is in-spec (800ns ± 150ns) or not, but if it is, it's extremely close. (This one fell on the edge of my sampler, so the results alternated between 0.9583µs and 0.9167µs; I took the average)
T1H: 0.6666µs -- this is probably actually 0.6250µs (aka 10 16MHz cycles). Easily in-spec (700ns ± 150ns)
T1L: 0.9583µs. Out-of-spec by 208.3ns (600ns ± 150ns)
I'm quite rusty in assembly and a complete newcomer to ARM. tolson, is there anything we can do with the existing code to get this in-spec?