In your Chicago example, the electoral college could easily magnify it.

Imagine, for a moment, that if the kind of voter fraud that Chicago earned a reputation for in the 1960s were to happen in a city like Miami nowadays. So let's say 5,000 fraudulent votes went to candidate A. In a popular vote system, that's unlikely to swing the verdict because that's such a tiny margin compared to the country's entire electorate. 5,000 is less than 0.004% of the number of people who voted in 2008.

But Florida historically polls close - easily close enough for 5,000 votes to swing which way the state goes. And this winner wouldn't just pick up that tiny sliver, they would get all 29 of Florida's electoral votes. That's a much bigger chunk of the pie - 5.4%. Worse yet, in the electoral college getting the votes is a zero-sum game. (Not so in the earlier case because the votes shouldn't have existed in the first place - so picking up a fraudulent vote for candidate A doesn't mean that candidate B lost a vote.) Candidate A getting 29 votes means candidate B loses 29 votes, so the real swing is more than 10%.

Meaning that in this hypothetical, the electoral college managed to magnify the influence of what was a relatively small instance of electoral fraud by a factor of more than 2,500%, into something that could easily swing an election.

Electoral fraud in Illinois would swing a maximum of Illinois's 20 electoral votes, or 29 in Florida

With a popular vote, it would be popular to have millions upon millions of fraudulent votes in a single place, say, Delaware, that would allow the opposing candidate to win even if there was supposed to be a landslide victory for the other candidate.

I think "Firewall" is a good analogy of this.

How, exactly, would one manage to organize millions upon millions of fraudulent voter registrations in a single place?

It would be better if your hypothetical example were at least remotely plausible. The entire population of Delaware is less than one million.

Keep in mind that my example of only 5,000 fraudulent registrations in a single place is already over an order of magnitude larger than notable examples from recent history.

Simply pick a state that uses electronic voting machines, and engineer Malware, or alternatively, hack the central repository.

I supposed millions in Delaware is a bad example, but it's certainly possible in a place like California.

"Take a jurisdiction like Chicago which is known for voter fraud and high turnout among the dead. "

An unproven case of a small number (<100) of possibly fraudulent votes in an election more than 60 years ago, and still going on about it as if it happens every election, and orders of magnitude worse.

We can have a better more granular "firewall". Instead of direct popular vote, simply quantize it into congressional districts instead of just the state level.

Who wins the district wins that EV. Who wins the state wins an additional 2 EVs (ie, aligned with senate numbers). Total would equal (currently) 538 EVs.

Some states already do something similar (see Nebraska, Maine).

Of course, this brings up the perennial problem of gerrymandering... which could be resolved by mandating perimeter to area ratios that keep things from looking too crazy.