Actually, I've explained a number of times why I don't like Truform. There are a number of reasons.
First of all, from a programming perspective (I'm sort of an off-and-on hobbyist programmer), there is no possible way that I could, for example, write code that would automatically make any model look good with truform enabled. It's just not possible. For this, you truly need a fully-generalized HOS technique available (It doesn't look like we'll see one for about 6-8 months in a consumer card, at the very least...though the rumored Primitive Processor that nVidia may be developing would be just what the doctor ordered). I had thought for a while that the RT-patches support in the GeForce3/4 cards would be general enough, but it seems that developers weren't so enthusiastic, and support has been dropped entirely (It was dropped in D3D a long time ago, and just recently dropped in OpenGL).
The second reason that I don't like Truform is that it always seems to produce undesirable effects on certain models. I'm of the opinion that even if it makes 99% of the game look better, but 1% of the game look worse, then it is not worth turning on. That is, a game only looks as good as its worst-looking part. I have yet to see sufficient evidence that any game has been able to satisfactorily enable Truform in every situation.
The third reason relates to the lifetime of Truform. Due to the hours required of the artists to modify their models for Truform, and also due to the increasing performance hits from increasing polycounts (See the UT2k3 Demo), it just seems obvious that Truform simply is not going to be viable going into the future.
As an aside, subdivision surfaces are not a totally useless concept, it's just that Truform uses the most basic and ugliest version of them. If we see more subdivision surfaces in the future, as opposed to other HOS techniques, hopefully they'll use much better techniques in generating the new surfaces. Actually, what I hope is that we both see a superior subdivision surface technique (that doesn't show the balooning effects of Truform...something that is most certainly possible), as well as a fully-generalized primitive processor. One thing to note, however, is that it is far from trivial to have backwards-compatibility with any sort of generalized HOS implementation for older hardware that does not support that particular HOS technique. Hopefully we'll see a standard technique arising soon...and, perhaps even more importantly, that that hardware implementation quickly filters down to the low-end.