Similar concerns have been raised by others. See, e.g., the talk "Bibliometric Evaluation of Computer Science - Problems and Pitfalls" by Friedemann Mattern (Institute for Pervasive Computing, Department of Computer Science, ETH Zurich).
There is also a very interesting joint report from three mathematical boards, which is definitely worth reading.
Finally, the "Sector Overview Report from the Computer Science and Informatics Sub-Panel (UoA 23)" after the British nationwide "Research Assessment Exercise 2008" (available at http://www.rae.ac.uk) includes the following passage:
We frequently found that citation counts were poorly correlated with the sub-panel’s assessment of the impact of the work examined. Citations also varied widely between research areas. For instance, much of the highly significant theoretical research, in which the UK is world leading, typically attracts low citation counts. Despite these low citations, the work is often found to have profound long-term impact on practical aspects of the field.
(The emphasis is mine and is not present in the original text.)
I hope that some of you will find these contributions interesting. I thank Catuscia Palamidessi and Vladimiro Sassone for pointing them out to me.