Improvements in  issue 886985  weren't enough, we ran out of capacity during issue 887241 upgrade:

https://ci.chromium.org/p/chromium/builders/luci.chromium.ci/Win10%20FYI%20Release%20%28NVIDIA%29/3669
https://ci.chromium.org/p/chromium/builders/luci.chromium.try/win-angle-rel/25

Logs for splitting are:
Running Swarming with args:
['query', '-S', 'chromium-swarm.appspot.com', '--limit', '0', '--json', 'c:\\b\\swarming\\w\\ir\\tmp\\t\\base_trigger_dimensionswnj_lc.json', 'bots/count?dimensions=gpu%3A10de%3A1cb3-23.21.13.8816&dimensions=os%3AWindows-10&dimensions=pool%3AChrome-GPU&is_dead=FALSE&quarantined=FALSE']
Bot config 0: {'available': 19, 'total': 21}
Running Swarming with args:
['query', '-S', 'chromium-swarm.appspot.com', '--limit', '0', '--json', 'c:\\b\\swarming\\w\\ir\\tmp\\t\\base_trigger_dimensionszn44e9.json', 'bots/count?dimensions=gpu%3A10de%3A1cb3-24.21.14.1195&dimensions=os%3AWindows-10&dimensions=pool%3AChrome-GPU&is_dead=FALSE&quarantined=FALSE']
Bot config 1: {'available': 19, 'total': 110}
Total bots: 131
Total bots after filtering: 131
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available
Chose bot config 0 because bots were available
Chose bot config 1 because bots were available

I guess what happens is:
1. All tasks are triggered
2. Only after that all tasks are run
Thus, bots can be available during triggering of all the tasks, but once the first task runs, bots are no longer available.
What I mean, is that logic in trigger_multiple_dimensions.py only works for first task, but the "available" status when triggering a second task would be different from "available" status when the second task needs to run.

I suggest we should ignore "available" status and just split according to "total".
Other options are possible, like waiting for the previous tasks to start running before triggering a task, or keeping history of what was scheduled in previous tasks.
Ken, could you take care of this?