As you say, there have not been many updates lately to Velvet/Oases, but there have been for Trinity. I have personally noticed better results with Trinity. Genes that were only partially assembled with Oases were fully assembled with Trinity.

This paper looks at particular genes for full-length reconstructions for phylogeny, and analyzes for chimeras. Oases doesn't perform so well. "Optimizing de novo assembly of short-read RNA-seq data for phylogenomics." http://www.ncbi.nlm.nih.gov/pubmed/23672450

I found this paper recently: Comparative analysis of de novo transcriptome assembly

My personal experience (multiple plant and arthropod species with >150 million reads) is that Trinity performance is superior to Oases multi-k assemblies, in particular when judged based on highly accurate (>99% identity) assembly of known transcripts.

Trinity in-silico normalization is also indispensable for large data sets.

The active maintenance of the Trinity package is also a huge plus.

Another player joins the competition. SOAPdenovo-trans

http://arxiv.org/abs/1305.6760