Is this a large number relative to the number of votes cast? Is it systematic across time and place, or is it an isolated occurrence? Did it, or can it, swing elections? A raw number of 900 is not sufficient information to answer these questions.
Also, I think from a policy perspective, one must compare the tradeoffs involved with the benefit of increasing the accuracy of the vote by X% to the cost of disenfranchising (or making it substantially difficult to vote) of X number of the people. For example, if this law increased accuracy by, say 2%, but imposed insurmountable obstacles for 10,000 people to vote, is the tradeoff worth it? (Note also that disenfranchising people skews the vote also, by a potentially greater amount than voter fraud, but then, that's the purpose, isn't it?)
I don't have the answer, other than to say that the social benefit from such legislation its not self-evident on its face, regardless of its partisan political benefit. (I care about social benefit, not partisan political benefit.)