@bored-art-40741:

It sounds like you do well internally on 6gb heap (what's your perm gen?) and with 4 workers?  And do you find yourself compiling "leafy" targets (with lots of potentially non-overlapping classpath dependency analysis) in parallel?

... we do occasionally see issues near very large leaves, but we hadn't investigated because retries in CI usually mask them. so yea, there are probably things to nail down there. one thing i wonder about is whether zinc actually needs transitive analysis, or just direct analysis...