There are several factors to consider:

Politics - Putting your project on a public forge can make it seem more credible.

Tooling - The quality of the tools available to the community can vary significantly between the public forges and the host-it-yourself ones. In some cases projects decide to use a hybrid approach which augments the tooling of the public forge - for instance using SourceForge for binary, source, and documentation downloads but hosting a JIRA installation for case tracking.

Cost - The economics are clearly different between the public forges and hosting one yourself.

Spin-off Projects - Part of the point of hosting your own forge is to host not only your own project, but to allow developers to create spin-off projects in the forge - e.g. for plug-ins or extensions to your project. Is this your intent?

Contribution Expectations - Some companies have ideas about which kinds of contributions are welcomed and which ones are not. These expectations might affect the choice of tooling and hosting.

Metrics - The quality of metrics provided by the forges differ also. What are your expectations when it comes to weblog and usage analysis of the traffic on the forge? Which tools meet those expectations?

You can always move the hosting at a later date, but you need to communicate how and why this is happening, and you may or may not lose some community along the way.

We have always hosted our projects on public forges augmented with JIRA and Confluence. We have looked at the options for hosting our own forge but have never gone ahead with it.

I am not sure if there is any really good reason to start a new forge. The existing sites making it very economical to quickly host a new project.

You might also consider hosting the project at an existing open source community, like Apache, Eclipse, OpenJDK, Mozilla, etc. Starting an open source project is more than just finding a hosting solution, you also need to think about the IP considerations, community building and development process. Most of these existing communities can help new projects with many of these issues.

the decision should be like UK financial services regulation (look how well it worked!).

That is, the natural decision should be to use a public infrastructure or existing org/community/platform structure - if the decision is to do something then that requires special justification. What is the return on private in this case from an IP perspective?

Also certainly worth looking at GitHub- more and more development and tree management is going on over there.

I think you need to ask what it is that would be gained by doing this yourself? Are you really an expert at hosting and running a forge type of site? What happens to your forge if your company disappears? Think about it from that perspective for a minute. Sun hosts a lot of things on their own but they will effectively disappear when the merger with Oracle goes through. What will happen to Sun's FOSS forge style sites such as NetBeans under Oracle's ownership? I for one would be more comfortable if these projects were hosted somewhere like SourceForge or github. Oracle could quite simply unplug them the day the deal goes through (not that I think they will but I am making an example). Now, if you are an Oracle, IBM, or Sun I might be more comfortable but if you are not a company like that I think developers would be a whole lot more hesitant to participate if you tried to roll your own. As the previous poster said - that is politics and credibility and they definitely play a roll.

My 2 cents is to find a public forge with the right tooling for your project and make a contribution to keep it running smoothly and host your project there.