I think that algorithms are definitely part of the process of reducing the time and costliness of creating 3D games. However, a warning against relying on them too much should be expressed.
SpeedTree isn't perfect. I was playing Oblivion only the other day, and I know that one big flaw I see is that many of the trees, at least in Cyrodil, are tall and old. Very few are short, or new. You see these great oak forests which have little to no undergrowth.
When I see a fully generated rainforest, which I'll recognise as realistic from my extensive experience of rainforests along Australia's Eastern Coast, with all four strata of emergents, canopy, understory, and forest floor generated in a way that none of the trees clip over each other, interacting properly, then I'll know that SpeedTree has done it. Until then, this system might be good for a start in making a nice forest, but I think that designers should tweak further than the process that SpeedTree provides.
What is especially exciting is that SpeedTree is only the tip of the iceberg. I expect that if the games industry knows what it's doing, there will be fast design systems like these for character morality, buildings, factional alignment, updating, and saving. One day, imagine if these systems could update themselves across all the games that use them, empowering modders and possibly designers to patch new graphical effects and environmental interactions into old games. Now there is a tantalising prospect.