Yesterday, I attended a lecture by Eugene Garfield, founder and chairman emeritus of ISI, in the Austrian State Archive. He talked a lot about "algorithmic historiography" and the software HistCite, which seem to be his current "hobby".
After one hour, one of the auditors interrupted the sermon and asked about the inadequacies and negative impacts of citation analysis, more precisely about the consequences of using SCI data for evaluating scientific staff. Mr Garfield (probably having been asked similar questions very often) said that he wouldn't be responsible for wrong handling of the SCI data and it would be better to have these sometimes imprecise data than to have none. This answer didn't really satisfy me...
But Garfield wrote a paper about the proper use of citation analysis for the evaluation of faculty that is worth reading. He says: "The ultimate evaluation involves an in-depth interpretation of each candidate's papers and books. The analysis should take into account the publishing and citing conventions of the field, the reasons why the candidate’s papers are cited, and adjustments for self -citations. I make these qualifications knowing that, in many instances, one can obtain important impressions about individual candidates by a mere glance at a five- or ten-year cumulation of SCI or SSCI. But this first crude impression needs to be qualified by the other impressions obtained through subjective peer review". The text can be obtained via Garfield's website as a pdf (part 1, part 2).