What about Gephi ?

I used it together with Bio4j for some gene-ontology visualizations and must say I'm quite happy so far. Here you have a couple of links with some samples:

http://blog.bio4j.com/2011/04/go-annotation-graph-visualizations-with-bio4j-go-tools-gephi-toolkit-sigma-project/

http://blog.ohnosequences.com/2011/03/playing-with-gephi-bio4j-and-go/

or try Cytoscape

Try GenMaPP.

You may like to try FunNet. There's a web server version and also an R package.

It takes a little effort to figure out how it works but basically, it does what you describe: takes expression data and maps it onto KEGG (as well as GO). See the tutorial for details.

The output can also be loaded into Cytoscape for visualization.

I didn't try the programs mentioned above so I don't know if they are more convenient than the solutions I'll suggest:

1. Ipath. It is only a web interface to display KEGG maps but the maps are customizable. I could not try it before and it is currently not working on my computer but their help page teaches how to customize the maps. The only drawback and that you need to first create a mapping file (Kegg Id to color) and then load it into Ipath. I agree this can be tedious if you want to do this several times.
2. R and KeggML (KGML, an XML format) files. With the KEGGgraph and KEGG.db libraries (part of Bioconductor) you can easily parse a KGML file (function parseKGML2Graph). This will return an object containing the nodes, edges,... You can draw this graphs and modify them, specifying colors for edges or nodes or changing the shape of the nodes for example. You can then simply specify the colours based on the gene expression level. I did this few months ago and it works quite well, one drawback is that it becomes a bit hard to read if there is many nodes/edges. This requires a bit of coding but it is quite well documented and might not need a lot of time if you are used to R.

Or try the Expression Analysis module of Reactome. To get an idea what is possible, have a look at the documentation.

You can also try ChiBE. It can load pathways from Pathway Commons and visualize it in an SBGN-like way. It also lets you overlay expression data on top of the nodes from a local file, from GEO or cBioPortal.

Here is the URL to the project page: http://code.google.com/p/chibe/