Starting with an scenario such as N test classes run:

com.parallelizable.one (duration 10 mins)
com.parallelizable.two (duration 10 mins)
com.notparallelizable.one (duration 1 mins)
com.notparallelizable.two (duration 1 mins)

I am interested in parallelizing classes with packaging com.parallelizable, ending with an pseudo-pipeline such as:

node {
test (com.notparallelizable)
splits = splitTests parallelism: 2
parallel ( "one": test(splits[0].list.join('\n')),
"two": test(splits[1].list.join('\n')))
}

My current workaround is filtering with grep:

parallel ( "one": test(splits[0].list.grep(~/com.parallelizable.*/).join('\n')),
"two": test(splits[1].list.grep(~/com.parallelizable.*/).join('\n')))

but obviously, the resulting tests wont be ballanced in number of elements. I was thinking in filtering at the plugin level, so the returned groups would be properly ballanced.

Im not sure if this improvement would be interesting for freestyle projects (i dont use them anymore), but for sure it will for pipelines.