Nice ideas, folks.   Currently, my code runs in ~20 ms (in my environment) when splitting a 61K varchar(max) value into 8 chunks, and at the moment, that's a bit less than half of what the best one appears to just do the split part in, so I'm pretty sure I'll stick with what I have, but perhaps the question to ask now is if anyone used a test rig on my code and measured performance across a rather larger number of values.   These values would, in the vast majority of cases, not be coming from column values in a table directly, but would instead be "assembled comma-separated lists" that end up being passed as parameters, and I suspect that could make a difference in how volume on this could be tested.