Hi Jonathan, If your sequencing budget is limited, and Transcript discovery is your priority, you could invest more in depth of sequencing and read length (>75Bp) rather than replicates. Also paired end data will be important. The assembly will only give you a snapshot of the input samples as well so if you think important transcripts are expressed in other tissues, conditions you may want to sequence those as well.

That being said, having replicates will allow you statistical power to quantify transcripts that are present in one sample/condition/tissue and not in another and look at differential transcript usage.

Here's a nice resource: http://rnaseq.uoregon.edu/#exp-design-experimental-objectives

I would argue that without replicates you'll likely have a lot of false positives to chase down with other experiments, and you'll probably also be missing a lot. Replicates are always a good idea. But as Jacob said you'll also want to look at your depth of sequencing as well. Its always a trade-off one way or the other.