[pfdietz]

If nothing else, eventually civilizations are confronted by questions (in math, say) that can only be solved by extremely large amounts of computation. We know there must ultimately be succinctly stateable problems that cannot be solved easily; this follows from the undecidability of math.

[varjag]

It could well be that harnessing the energy of one star is sufficient for most of that. This is very hard to detect for an observer.

Going K3 to 100 billion stars of the galaxy "only" gives you a 10^11 constant factor speedup at enormous expense and with Amdahl's penalty at galactic scales.