Summary: decriminalisation clearly doesn't lead to catastrophe, seems to reduce harm and might reduce consumption in some cases
Before attempting a proper answer to the question as asked it is worth making a clarification: the expected benefits of decriminalisation are much broader than just a reduction in use (lower crime and lower policing costs could happen even if consumption increased). Asking whether consumption decreases is a very narrow question compared to asking whether harm as a whole is reduced. I'm sticking with the narrow question here and assuming the broader question might get asked later so other evidence can be brought to bear.
It is also worth defining what we mean by decriminalisation. I think there is evidence about 3 different models (there are probably others but these cover a broad range). I'm going to call them the Swiss model, the Dutch model and the Portugese model.
In the Swiss model (which applied to injected opiates) the point is to medicalise the problem of addiction by providing medical grade opiates and clean needles to addicts in a supervised environment. The intent is to reduce the harm from criminality (i.e. stealing to pay for the drugs), medical complications (impure drugs of uncertain concentration, dirty needles etc.). Addicts also have access to the services they need for withdrawal. The model has been so successful in Switzerland that even the relatively prohibitionist UK is piloting it (Lancet report here). But assessments have been done on the number of regular addicts. And the results suggest that the numbers of long term addicts drop albeit slowly (broader measures of harm reduce by large amounts).
The Dutch model represents a sort of supervised tolerance of soft drugs (mainly cannabinoids) from licensed premises. The detailed rules have been tweaked several times in the last 40 years. But there is little clear evidence on the relationship between the specific rules and consumption. One expert argues in this debate thread in the BMJ (see the contribution by Rolles) that:
It should also be noted that cannabis use has risen across the Western world since 1976, but has risen slower in Netherlands than many other countries. Young people's cannabis use in the Netherlands is substantially lower than either the UK or the US.
A more thorough study in 2002 concluded that:
Although changes in cannabis policy went along with changes in availability of cannabis and prevalence of cannabis use, it is questionable whether changes in cannabis policy were causally related to trends in cannabis use. Cannabis use also developed in waves in other European countries that did not decriminalize cannabis, as well as in the US. Consequently, trends in cannabis use seem to develop rather independently of cannabis policy.
So while it is clear that the jeremiads of anti prohibitionists who forecast an explosion of use on decriminalisation have not occurred, we can't say that, in this case, consumption actually declined.
The third model is Portugal. Here possession of up to ten days supply was decriminalised and addicts referred to treatment if caught. This summary of the results claimed a significant reduction in use of injectable drugs after the policy had been in action for some time. It should be pointed out that these results are controversial (see the comments on the report here.)
My summary of the evidence above is that it clearly shows that decriminalisation is not a catastrophe leading to big increases in the consumption of dangerous substances, and there is some evidence that consumption might fall. In all cases it is likely that harm was substantially reduced.
There is plenty of other debate on this topic (thought the quality often leaves something to be desired as many of the participants seem unable to separate logic end evidence from their emotional and intuitive starting positions). A good place to start is the BMJ debate here but the follow up online debate is also worth reading.
A very recent publication from an international group, the Global Commision on Drugs is available here but it is only fair to report that some of their numbers have been criticised (see the analysis by StraightStatistics here).
EDIT. A recent summary of the Portugese case appears in the BMJ in october 2011 and reinforces the above conclusions.