I'll take this in two sections: comments about the web app and comments about the manuscript.
Web App This is a nice web-app that attempts a difficult job. It is great that it is presented in multiple forms, including standalone and VMs.
Often people do not like the idea that their email will be taken, even if it does mean that they can retrieve results later. A less intrusive mechanism is to provide a job id on submission which can be used to get at results, either through a form or a URL.
The help mouse overs on the main page do not work.
Manuscript The language describing what a reciprocal BLAST can achieve is inconsistent. Although it is clear later in the manuscript you are aware that a RB can only find putative orthologues, it should be clearly stated throughout. At some point you might want to clarify how orthologue detection requires strong phylogeny.
The manuscript as written kind of makes a trap for itself. It tries to claim that RBH and this tool will help large scale analysis where only small scale analysis could be done before, but in doing so it invites us to consider RBH in a new (ish) mode - that of big data tool. And if we are to use it there, well, we need to know how well it does in that domain. In this case then this manuscript must do that large scale comparison and sadly, the manuscript is a bit lacking.
Considering RBH as a method for big data then, I don't think the utility of the method and the tool is sufficiently well tested. A glaring omission is that there is no attempt to describe the error of the method, that is the number of false putative ortholgues or any attempt to develop a metric for the believability of the whole set of data. For a tool whose main selling point is that it can find broad patterns in large datasets, then it really needs some measure of how many of the putative orthologues it classifies are right or wrong. The experiment carried out is simply a run of the tool. As a minimum I'd hope that you'd compile a list of known orthologues curated carefully and manually (there are lots of databases with these, (orthologene.org, orthoMCL, compara @ EMBL) , then run your tool and assess how many of these you'd found. Until a true benchmark experiment is done, then the manuscript lacks a sufficient demonstration of the tool's utility.
If we aren't supposed to consider this as a new big data tool and just a useful implementation of RBH, then the language and claims about being able to compare gross patterns across large phylogenetic distances need to be toned down in the manuscript.