I definitely agree here - the inconsistency that makes it less valuable to me on offence is exactly what makes me avoid it on defence (and in Delves, too). It may not always work, but it does have the potential to go horribly wrong.
I think there may be some merit in making transformations less expensive mana-wise, but also less devastating, and able to be countered, which could help bring it slightly back towards the strategy (rather than slot machine) side of gameplay.
Specifically with Dragon Eye, I could perhaps even be happy with it costing less mana, but keeping the Baby Dragon at max level/traits, as well as the Dragon it can then transform into (giving you a bit of a chance to fight back, but still breaking up the strategy/rhythm of your match). Maybe have it destroy proportionally fewer gems, as well.
On the broader scale of things, what I think really needs to happen (not to say that Transformations can’t or shouldn’t be revisited) is for there to be greater rewards for setting and defeating good PvP defences.
Try this:
-
Each time your defence team beats an opponent, add a certain amount of reward (e.g. 100 gold) to your defence ‘pool’ and compound it, similarly to how tributes (and defence rewards?) used to be calculated.
-
When someone beats your defence team, they earn the contents of the ‘pool’ in addition to the normal rewards, as do you (
Double Jackpot?!?!). This is your reward for contributing to the dynamic, challenging and enjoyable PvP environment.
This way, there would actually be incentive proportional to the variable difficulty of teams rather than solely based on team power.
I liked this one 
, and/or
although that one’s a bit more like your traditional Polymorph/turn an opponent into a chicken for a few turns, which could almost be its own thing.