My point is that it doesn't need to be "worse" or overcost. It's fine if it provides an advantage compared to the regular heroes you have. There's only 2 major caveats (aside from having the point costs be balanced of course)
1) Don't provide a unit type that they normally don't have (e.g. don't give Khorne wizards and don't give Fyreslayers mobile or ranged stuff)
2) Be very carefull with the types of abilities you allow them to have so you don't accidently end up giving them broken synergy with the rest of the army. E.g. don't give the necromancer hero a super powerfull, but difficult spell, and then be surprised when he's used as a vehicle purely to give that spell to Nagash.
Sticking a starpriest on a ripperdactyl, although a significant advantage thanks to flying, should be more than fine. Hell, mechanicly you can already do that with the cloak of feathers. All that's needed is just to make the starpriest say a 130-150 points instead of a 120. Even the added combat prowess shouldn't raise any issues at that pricepoint. Similarly, a starpriest with an extra wound or two, or a 4+ save instead of the puny 5+ shouldn't be gamebreaking provided it comes with a slight bump in pointcost.
In contrast, sticking a skink on a flying monster who can easily beat a terrorgheist into submission probably would be problematic as that means giving us a monster that far outclasses everything else we have. And chances are that's going to be too much when combined with all our potential buffs, even if it is appropriatly costed at 400 points.
Allowing freedom definitly makes things more complicated. But they seem to manage with the loadouts in 40K. So it's not impossible.
Unless you have an absolutly godlike memory I'd find that questionable. Don't get me wrong, I'm sure you can remember "X will annihilate MSU units of skinks" or "Y should be tarpitted by ~20 bodies, at least for a turn or two". But beyond that I find it unlikely you know the exact output of every unit in the game, let alone when also having to take into account potential synergies of what's on the table & damage tables. It always remains ballpark figures, and for that rules of thumb like "big healthy monster will do loads of damage, wounded monster will do somewhat less damage" and "hordes of 40 models will stab me a lot" should already give you enough to work with.
Having said that, these heroes do need to fall within certain bandwiths. A custom hero on a behemoth should do say an average of 20-30 wounds in combat against a 5+ save, with a melee focused hero being on the high end and a wizard focused hero on the low end. Not have a wildly insane bandwith like 0-30 depending on the exact loadout (fyi, numbers are completly arbitrary). If that bandwidth gets too wide you do indeed get the problem you've mentioned.
fyi, the most expensive behemoth I can think of is one of those Ghoul kings on terrorgheist at 440 points. At least if you ignore forgeworld stuff. So you can still get something top tier.
