The best question to ask when contemplating optimization is "How often is this called?" If a doesn't call b often -- once an hour is not often -- then there's probably no reason to include b in a.

Another question to ask would be "How does this affect the user?" If a is an interactive program, and the user must wait for b to launch, it might improve the user experience to include a into b. If a is not interactive, b could still affect the user if it uses an excessive amount of CPU or IO at startup.