I guess one option is the indirect route - by shutting down or at least decreasing the activity of host proteins and pathways the virus requires:

http://www.pnas.org/content/111/40/14613.short

The other option you mention, acting on viral mRNA, could be a product of virus-host co-evolution I guess. Let's say a certain plant has an RNAi system that - initially by chance or as a side effect of recognising similar things - recognises a certain viral mRNA. Upon infection, this would be an advantageous thing to have and the corresponding mutations in proteins (and nucleic acids) of the RNAi system would get fixed...but that's a bit speculative and coarse, not sure what the literature says about that. However,  it's safe to say that much: it will be down to evolution :)

Also, I think the issue has links to the general mechanisms with which cells detect non-self nucleic acid molecules (in the cytosol) in general - in many cases that works without any RNAi involved (but may in turn be counteracted by the pathogen, see e.g. pox viruses)

Sorry, not an expert for this at all, I just found the question interesting and gave it a shot. Other people will know better.

Hey Danny,

I am very interested in your question. I personally have worked in strasbourg at antiviral immune responses in Drosophila melanogaster. Fruitflies rely for their resistance on siRNA. I was wondering what you question is referring to? Do you say that exogenous siRNAs target also host protein expression?

Well, in most cases you have an exogenous RNA which is inside the cells (e.g. coming from a RNA virus infection) which starts ampifying. A RNAseIII (e.g. Dicer-2 in Drosophila) will recognize this exogenous nucleic acid and process it and hand it over to an Argonaut protein (e.g. Ago2 in Droso) which then takes part of the silencing complex and starts the cleavage of the exogenous RNA such as the viral genome or the viral mRNA. This restricts the viral amplification.

In the case of siRNAs which regulate the expression of some endogenous genes. I haven't read anything like that before. I would be interested to see your article you are referring to then.

But I could imagine some cases which could result in an influence of host gene expression by exogenous-derived siRNAs. Some genes in our genome are actually derived from viruses therefore there might be a cross-reactive part.

Furthermore siRNA are only 21 nt long. It is not very likely but there might be cross-reactions because of some overlapping sequence parts. 

However it is still hard to tell how siRNAs are actually recognized and which molecular pattern decide the cleavage of the exogenous RNA. This makes it all a bit difficult.

I would be glad to expand more on this, but I really hope this might have helped you already a bit. So feel free to contact me. However, I also attached a nice review on anti-viral immune responses in Droso from my lab. Have a look if you like... (it also shows the similiarities to human IFN expression)

Kemp, C., & Imler, J.-L. (2009). Antiviral immunity in drosophila. Current Opinion in Immunology, 21(1), 3–9. doi:10.1016/j.coi.2009.01.007

si RNA can block Viral mRNA by binding due to its complementority. mRNA synthesis can be by passed by some positive RNA viruses like dengue but it cause the pathogenesis.

Thank you all for your helpful answers.

@Felix: yes, as you said, I wondered about the case in which the exogenous siRNA could target the host genome. Briefly, after the exogenous siRNA is recognized and cleaved by Dicer,and is incorporated into other proteins to form a complex (RISC and AGO2 as in your explanation and also in this photo). This complex then binds to the mRNA molecule, but what happens if this mRNA is transcribed from DNA of the host cell. May this interfere the expression of the host genome? I'm not really working in this field, I just found it interesting when reading some books related to my major. Anyway, I'm also interested to further studies about this. Again, thanks for your explanation.

@ Danny, hi again. I am really sorry for the late response. I am glad that I could help you. Prinicipally, Surendra is right with his answer, however, you should see that siRNA are fragment of 21 nt, which make them as matter of facts really highly specific for the invading (for instance) viral genome and should not disturb with the host genome. The siRNA system is generally based on a pretty much 100% complementary meaning every single nucleotide should match in order to introduce the cleavage. You can calculate by yourself how many different 4^21 possibilites there are to encode a siRNA. Therefore the targeting of host genome expression is rather small using the siRNA-pathway. I hope this could help.

I most certainly are with you, Danny, the field of siRNA or generally RNAi is very much interesting and if you are really keen of this field I would suggest you to read up on long non-coding RNAs (lncRNAs) which are the upcoming field of gene regulation via RNA which is still only poorly understood but makes up an immense mass of RNAs produced in human cells for instance.