I've finished writing up a script to allow one to perform slightly more standards-compliant SPARQL on PHP. RAP, the RDF API for PHP, supports SPARQL, but doesn't support specifying FROM statements. 
There's an easy way of sorting that though - just take your SPARQL query, grep it for "FROM (.*)", and take each of the addresses it returns, load them in to RAP's memory model, and then run SPARQL across the model. My code doesn't yet support FROM NAMED, but it'd be trivially easy to add along the same lines that I have already suggested.
The problem is that you can't easily test against the GRAPH - since it's all just chucked together. The idea of this kind of hacking isn't to make it so that people can build a gigantic, totally compliant system. There are systems like Jena and ARQ (both Java-based) that do this (RAP also has a way of using MySQL to store large RDF datasets and run SPARQL over them).
But if the SemWeb is going to take off, it's not going to be application developers writing huge industry-strength Java servers, it's going to be individuals that are experimenting. Once they realise the benefits that the SemWeb brings, they can then go and evangelise those benefits, and then the big companies can pick up the Jena-type libraries and build industrial-strength services.
If you had to be the size of Google in order to develop web applications, Google would not exist. 
I wanted to launch a little semantic product that uses OPML and Grazr as an output format and display mechanism. I've been trying to get it going since Christmas Eve, and I'm still not there. Hopefully, I can finish it off this evening and launch it properly tomorrow.
Just to give you an analogy, imagine if you had a collection of menus from lots of different restaurants and a menu reader, you take all of the menus, extract what they are saying, plonk them on to one giant menu and then ask that giant menu questions like "show me all the restaurants that are open after 5pm that serve vegetarian pizza". You might be able to get Google to give you that information. If you add "the owner must be interested in karate or the Gillmor Gang" to the end of that query (insane though it may be), then you need to use SemWeb technology like SPARQL.
I wrote up yesterday (but didn't publish) that the Semantic Web needs to be simple enough that anyone who can install WordPress can get going with it - not just as a 'producer' of RDF content, but as a consumer of RDF and a producer of services that crunch through RDF content. It's taken me all of about three days of trying out different pieces of software and reading documentation to get to this point. SemWeb needs to get easier if it's going to get widespread adoption.
Compare this to XML. If someone chucks me an XML file, I can crunch through it relatively easily with tools like the DOM and SimpleXML parsers. I can turn it all in to an array and crunch through it. It is reasonably easy to read XML and do things with it. Reading RDF triples and doing things with them needs to be just as easy.
We've got a lot of work to do if this is going to take off. I'm going to record a podcast later with some more reflections on the SemWeb and the process of getting in to it.
