Yeah, that's an interesting question. I think it depends on a couple things re: how the docker provider handles caching and gc. I'd say our average images are around 10gb, with some of them nearing 90-100gb at the moment. The builds happen in parallel, so some layers may be duplicated before they're added to the local cache, which definitely complicates things. I also opened this issue:
https://github.com/pulumi/pulumi-cloud-requests/issues/381
if that's an easier place to discuss. Happy to follow up wherever is best for you!