I ran some numbers and I think it would make sense to be smarter about component build scheduling, building the largest components first and filling the available queues (koji biulders / local threads) with the small ones.

I recorded the local build times for all the Base Runtime components and compared the best v. the worst case scheduling scenarios for various number of build queues. The average time saved was around 20% (nearly 40% in the very best case). Of course the results depend on the packages you're building, the number of components, the number of queues and your hardware specs but it's always going to be non-zero.

A few more notes:

• you could gather your own build time statistics or you could get this information from koji; look at the past, say, three successful builds of the component and get the average build time
• smart scheduling should be done per build order group
• it should also be done after compoent re-use is evaluated, counting the re-used components as taking zero or close to zero time
• this will affect both local and koji builds but will be more noticable locally