If I understand correctly, there is no built-in way to cancel an F.delay { ... } block.

Well, the idiomatic thing to do is to split it into multiple blocks, so that there is possibility for cancellation between them.

But, there are bigger concerns here than cancellation.

Is this actually "better" than doing F.interruptible { ... }? Or does this have a downside?

Yes, there is a downside: your longComputation() is essentially a blocking operation so running it in F.delay it will take over one of the WorkStealingThreadPool's limited threads, absolutely no different than if you did blocking I/O and thus has all the same caveats:

running blocking code on our compute pool is very bad. If we're running on a node with 2 CPUs and we evaluate a blocking call like IO(Source.fromFile(path).getLines()) then for the duration of that operation our capacity to evaluate IO fibers is halved. Run two such operations at the same time and your application effectively stops until one of those blocking calls completes.

From https://typelevel.org/cats-effect/docs/thread-model#thread-blocking

This is why F.blocking and F.interruptible shift the work to a new threadpool, specifically to handle blocking operations without starving the primary compute pool.