Lynne Featherstone has been blogging about the fact that the prime minister currently gets to pick the election date (and notes that Gordon Brown was for fixed parliamentary terms in 1992, but not, wonder of wonders, now that he's in charge).
On the one hand, a change to fixed parliamentary terms ensures that governments can't wait until the high-water mark and call an election when they are popular. On the other hand, considering the fickleness of the media and of opinion polls, having fixed terms could lead to a situation in which a government that had been largely popular throughout most of its rein was thrown out at a low point.
This strikes me as a sampling problem. Imagine a graph with government popularity on the Y axis and time on the X axis - you'd get a wiggly, wavy line. Taking a sample every 4-5 years is a very, very imprecise way of recording how popular the government has been over time.
I propose that government is sampled more often, and that, say, 20% of constituencies go to election each year, at a neutral time of year like spring or early autumn. That way, a government's power can be continually regulated by the people. Of course, there would be no more landslide victories for one party or another, since it would take several years for the total composition of parliament to change. But would that be a bad thing?