IgG is rarely used as a control these days because it turns out to be very problematic to use (due to getting very very very low amounts of DNA from the IP). In general, it's best to use a straight input sample (i.e., something that wasn't ChIPed) or H3. I'm not a huge fan of using another IP as a control, since then you're relying on multiple antibodies, possibly with different IP efficiencies. Having said that, if the H3K4me3 antibody isn't completely specific to the modification, then using H3 as an input will certainly help to remove non-specific H3 peaks.