OK, yet another crazy idea of mine.
Generally, we we coerce to classical SVD form with singular values, then Tikhonov regularization can be probably optimized post-decomposition. Indeed, i can see no reason why we can't control the smoothing at the prediction stage by hacking the predictor as in following. Consequently, if we can, then we can also optimize degree of smoothiness on hold out data after decomposition is done, perhaps even cross-fold. but we don't have to rerun it again and again. Finally, a hack for ALS-WR is enclosed too, but in this case this is more intuitive than derived. i guess i can try it out in R. It is still possible that it is a complete nonsense of course. https://docs.google.com/open?id=0B883AxfQlYWANDllNWQ1ZDQtOTEzOS00MWM3LWI4MjItNDQ3MDg2ZWMzMmE3