So, ignoring wind for a moment and talking only about hills... BestBikeSplit demonstrates that a totally flat power output (VI=1.00) is not the fastest way to race a course. However, this brings into play many other variables, like what is a good target VI, what's a good target max power (and for how long) above the goal average power, etc. I use BestBikeSplit to simulate courses for upcoming races. I model a race in BBS and load the profile in Zwift and ride it, then see how I feel when running afterward. I'll usually go through 3-4 revisions of the race plan, playing with target AP, IF, VI, max. power, etc. I'll then use that power plan on race day. Having done this a few times now, I can say that if you haven't done this sort of analysis before race day and you think you're going to just "go a little harder on the hills and compensate on the downhills" you'll almost certainly overcook the bike. In that case I'd recommend just holding a flat average power as best as you can.