A couple of months ago I suggested jacking up property taxes on downtown lots while cutting property taxes on buildings in order to discourage "land banking." (But see here for my characteristic wishy-washiness.)
Market Urbanism thinks this is a bad idea because it would discourage land speculation. He argues that, while land speculation might stick us with surface parking lots in the short run, speculation serves a useful function by reserving land until development is optimal for the site. This ultimately facilitates denser development. (He offers a couple of other reasons as well; check out his post.)
I'm on firmer ground, though. Although I couched my suggestion as a way to penalize vacant lots, taxing land rather than buildings is perfectly sound economics. A tax on buildings is a tax on capital. Taxing capital discourages investment in capital. This underinvestment creates a deadweight loss: the investor foregoes the return the capital investment would have yielded; consumers (potential renters) lose the consumer surplus they would have enjoyed. Taxpayers are no better off because the government can't collect a tax on a building that isn't built. Switching to a land tax eliminates this deadweight loss.
By reducing the returns to investment, a capital tax makes undeveloped lots (such as surface parking lots) relatively more attractive and causes investors to demand higher returns from their investment -- which may indeed mean greater density. But that assumes property values eventually rise enough to create those higher returns. That's no sure thing.
In any event, who's to say a 2.5% tax encourages an optimal level of density? Why not a 4% or 8% tax on buildings? Or 20%? I don't see a rationale for choosing one rate as "optimal." It's better to let investors make decisions based on the real returns to investment and avoid the distortionary effect of taxes altogether.