My personal preference ( as you may have noticed at ANY23-19 ) is to stay with the OpenRDF API, even if you don't use the Sesame implementation of it. All of the elements of the OpenRDF API are specified as interfaces and there are no hidden static singletons in the background that you can't reimplement another way when you need to. In addition, you can pull in the API using Maven without having to use the implementations which I view as a large point in its favour.
I made quite a few changes yesterday to basically suit my use cases. These including fixing the cause for this issue by switching to RDFFormat as the central point of reference for the utils classes. I pushed the changes to GitHub at https://github.com/ansell/any23 . I am not sure how many of the changes you will want to pick up but you may find something interesting. The most visible change is that I split the core module up into a large number of smaller modules. You probably won't want to do that just yet, but I found that I could work with it and visualise it more easily if everything wasn't in a single core module. It may seem a little convoluted, particularly having the test resources in their own module, separate from the core unit tests, but it makes it simpler to use the test resources in maven to unit test the other modules, without having to either split up the test resource bundle or replicate some resources in other modules.
The other very visible change is that I deleted the Any23 NQuads library in favour of importing the MIT licensed Sesametools NQuads library ( https://github.com/joshsh/sesametools ). It contains both a NQuadsParserFactory and an NQuadsWriterFactory that make it work with the static Rio methods that are all based on Service Provider Interface (ie, META-INF/services), including Rio.createParser() and Rio.createWriter(). This meant that the use of RDFFormat was no longer a breaking point, as it was when I first tried to switch and I was wondering why Rio.getParserFormatForFileName() was not working for NQuads. It turned out to be due to the missing ParserFactory in the Any23 NQuads implementation. In either case it doesn't really matter for long as there are plans inside Aduna to pick parts from one or both of the Any23 or Sesametools NQuads libraries (or the other two open source OpenRDF NQuads implementations) and implement them internally in a new sesame-rio-nquads package.