The ethical issues of accepting nonzero fraud are that striving for zero fraud creates program design changes that lock people out of benefits. If you design a health care system that aims for 0% fraud, some measurable number of people are going to be deprived of care because the registration and billing procedures are too onerous. With taxes, aiming for 0% noncompliance will prevent people from taking advantage of deductions and credits.
This isn't hypothetical; it's the issue underling the "program design" controversies about means-testing in public policy.
Not to mention that enforcement has rapidly diminishing returns. Even if your only goal was to maximize tax revenue (minus cost of tax administration), and you didn't care at all about people being able to take advantage of deductions, the optimal amount of fraud is almost certainly non-zero.
(And of course, if you did want to maximize tax revenue, you'd focus enforcement on the big fish.)
This isn't hypothetical; it's the issue underling the "program design" controversies about means-testing in public policy.