The random transposition shuffle on repeated cards induces a Markov chain on the quotient space of arrangements with multiplicities, and is equivalent to the many-urn mean-field Bernoulli-Laplace model introduced by Scarabotti. Writing $n=ml$, where there are $m$ card types and each type appears $l$ times, we determine the limiting profile for the total variation distance to stationarity at times $t=\frac{n}{2}\left(\log n-\frac{1}{2}\log l+c\right)$, under the assumption $l=ω(1)$. Scarabotti previously established that this process exhibits cutoff at time $\frac{n}{2}(\log n-\frac{1}{2}\log l)$; our result refines this by identifying the precise asymptotic shape of convergence inside the cutoff window. We show that the limiting profile is asymptotically Gaussian, with different explicit forms in the regimes $m$ fixed and $m=ω(1)$. Together with our previous work on the fixed-$l$ regime, where the limiting profile is of Poisson type, this yields the cutoff profile for the random transposition shuffle on $n=ml$ repeated cards for the full range of parameters $m$ and $l$. Our argument has two main steps. First, we combine Scarabotti's Fourier-analytic framework for the many-urn Bernoulli-Laplace model with the approximation method of Jain-Sawhney (arXiv:2410.23944). More precisely, we compare the original shuffling measure with an explicitly tractable auxiliary measure directly on the repeated card quotient space, rather than passing through an intermediate comparison on the full symmetric group; this step relies in particular on our new estimates for Kostka numbers. Second, we reduce the limiting-profile problem to quotient fixed-point statistics and analyze them via Hoeffding-type combinatorial central limit theorems.