I think, {lo ca broda cu zmadu lo pu brode} does not say what you want to say.
The point of sumti is that they represent some referents. For simplicity, I will use sumti that have exactly one referent each. Usually a sumti reveals some information about its referent, so, using that information, the listener can correctly guess the referent of each sumti. For example, {lo blabi ku} says that the referent is white. That information, however, does not have a direct relation to the truth value of the main bridi, which the sumti is a part of. It does not matter how we refer, only what we refer to.
Let's consider an example: {lo pu verba ku ca ciska dei}. In {pu verba}, {pu} is a part of the selbri. {lo pu verba ku} says that the referent was a child. It does not say anything about the state the referent is at the time of the main bridi, which is now. Maybe he is still a child, maybe not. So, {lo pu verba ku ca ciska dei} does not say that a child writes the sentence. Actually, I wrote the sentence, so the writer is an adult; but some time ago he was a child, for sure.
Back to your example, I would interpret it as "something that is/does broda exceeds something that was/did brode". If you put a time tense into the main bridi, I will take it as if comparison of both objects happen at that time, regardless of tenses of inner selbri.
{do pe pu} will not work either, if you talk to one person. If base sumti has exactly one referent, no relative clause can change that referent; and in the aspect of zmadu3, {do pe pu} will be exactly the same as {do}... Or I may be terribly wrong in this regard, and {pe pu} magically sends your listener to the past. :D
I think, what we need is a reference not to the object itself, but to a state of that object at a certain point in time. A natural tool for that task is abstraction. So, {lo ni ko'a ca brodi cu zmadu lo ni ko'e pu brodi} does work. But
{lo ni ko'a [ca] broda cu zmadu lo ni ko'e pu brode}
does not, unless {broda} and {brode} are the same. Maybe it's a typo.
And by the way, to make the intended sense, not only "types" should
match, but scales. For a more explicit version:
{lo ni ko'a ca brodi kei
be zu'i goi ko'i cu zmadu lo ni ko'e pu brodi kei be ko'i lo ka ce'u
barda}
Since zmadu1 and zmadu2 are numbers in this case, zmadu3 should
be "is big". The original property that we compare {ka ce'u brodi} goes
inside {ni ... kei}, and {ko'i} is some typical scale to measure brodi-ness.