[AndMetal]

Here's the code:

zone/attack.cpp

	if (damage > 0)
	{
		// Give the opportunity to throw back a defensive proc, if we are successful in affecting damage on our target
		other->TriggerDefensiveProcs(this);
		
        return true;
	}
	else
		return false;

zone/mob.cpp

void Mob::TriggerDefensiveProcs(Mob *on)
{
	if (this->HasDefensiveProcs()) {
		this->TryDefensiveProc(on);
	}

	return;
}

zone/attack.cpp

bool Mob::HasDefensiveProcs() const
{
	for (int i = 0; i < MAX_PROCS; i++)
        if (DefensiveProcs[i].spellID != SPELL_UNKNOWN)
            return true;
    return false;
}

zone/attack.cpp

void Mob::TryDefensiveProc(Mob *on) {
	// this should have already been checked, but just in case...
	if (!this->HasDefensiveProcs())
		return;

	if (!on) {
		SetTarget(NULL);
		LogFile->write(EQEMuLog::Error, "A null Mob object was passed to Mob::TryDefensiveProc for evaluation!");
		return;
	}

	// iterate through our defensive procs and try each of them
	for (int i = 0; i < MAX_PROCS; i++) {
		if (DefensiveProcs[i].spellID != SPELL_UNKNOWN &&
			IsValidSpell(DefensiveProcs[i].spellID)) {
				if (MakeRandomInt(0, 100) < MakeRandomInt(0, 20)) {
					ExecWeaponProc(DefensiveProcs[i].spellID, on);
				}
		}
	}

	return;
}

zone/mob.cpp

void Mob::ExecWeaponProc(uint16 spell_id, Mob *on) {
	// Trumpcard: Changed proc targets to look up based on the spells goodEffect flag.
	// This should work for the majority of weapons.
	if(spell_id == SPELL_UNKNOWN)
		return;
	if ( IsBeneficialSpell(spell_id) )
		SpellFinished(spell_id, this, 10, 0);
	else if(!(on->IsClient() && on->CastToClient()->dead))	//dont proc on dead clients
		SpellFinished(spell_id, on, 10, 0);
}

From what I can see, it looks like the only thing that affects the probability of a reverse proc proccing (and maybe a woodchuck chucking?) is the random number generator. I must say though, creating 2 random numbers could create some funny odds of proccing, which may be the issue. I had to create a little program to calculate the probability, but using a sample size of 100,000,000 tries consistently gives a probability to proc of about 9.91% (specifically 9912469/100000000), at least on Windows:

#include <iostream>
using namespace std;

int MakeRandomInt(int low, int high) {
	if(low >= high)
		return(low);
	return (rand()%(high-low+1) + (low));
}

int main() {
	int l1 = 0;
	int h1 = 100;
	int l2 = 0;
	int h2 = 20;
	cout << "MakeRandomInt(" << l1 << ", " << h1 << ") < MakeRandomInt(" << l2 << ", " << h2 << ")" << endl;
	int tries = 100000000;	//100,000,000
	int count_true = 0;
	cout << "Running";
	for (int i = 0; i < tries; i++) {
		if ((i % 1000000) == 0)
			cout << ".";
		if (MakeRandomInt(l1, h1) < MakeRandomInt(l2, h2))
			count_true++;
	}
	cout << endl;
	int count_false = tries - count_true;
	float percent_true = ((float)count_true / (float)tries) * 100;
	float percent_false = ((float)count_false / (float)tries) * 100;
	cout << "After " << tries << " tries, " << count_true << " (" << percent_true << "%) were true & " << count_false << " (" << percent_false << "%) were false." << endl;