ProtoValuePtr was specifically designed to share actual content of proto fields in EntryKernel. These fields are identical 99% of the time so the current approach would result in significant overestimation by as much as a few KB per EntryKernel instance.

The code below it seems just adds size of all specifics_fields elements together which in most cases would result in doubling (and potentially even tripling) the estimated size.

size_t EntryKernel::EstimateMemoryUsage() const {
  if (memory_usage_ == kMemoryUsageUnknown) {
    using base::trace_event::EstimateMemoryUsage;
    memory_usage_ = EstimateMemoryUsage(string_fields) +
                    EstimateMemoryUsage(specifics_fields) +  <----- Here
                    EstimateMemoryUsage(id_fields) +
                    EstimateMemoryUsage(unique_position_fields) +
                    EstimateMemoryUsage(attachment_metadata_fields); <----- Here
  }
  return memory_usage_;
}

A better approach is to compare the underlying value address for equality. It is OK to test consecutive elements only e.g. specifics_fields[0] to specifics_fields[1] or specifics_fields[1] to specifics_fields[2].

The same applies to attachment_metadata_fields.