Why does the Quartus fitter fail to pack any ALMs when it does not find a fit?

I have a design which is on the limit in terms of FPGA logic utilization. I've noticed that when the fitter fails to find a fit, the "[B] Estimate of ALMs recoverable by dense packing" component of the ALMs needed calculation is 0. When the fitter is able to find a fit, some ALMs are able to be recovered.

+-------------------------------------------------------------------------------------------+
; Fitter Resource Usage Summary                                                             ;
+-------------------------------------------------------------+---------------------+-------+
; Resource                                                    ; Usage               ; %     ;
+-------------------------------------------------------------+---------------------+-------+
; Logic utilization (ALMs needed / total ALMs on device)      ; 32,157 / 32,070     ; 100 % ;
; ALMs needed [=A-B+C]                                        ; 32,157              ;       ;
;     [A] ALMs used in final placement [=a+b+c+d]             ; 30,284 / 32,070     ; 94 %  ;
;         [a] ALMs used for LUT logic and registers           ; 13,292              ;       ;
;         [b] ALMs used for LUT logic                         ; 11,471              ;       ;
;         [c] ALMs used for registers                         ; 5,521               ;       ;
;         [d] ALMs used for memory (up to half of total ALMs) ; 0                   ;       ;
;     [B] Estimate of ALMs recoverable by dense packing       ; 0 / 32,070          ; 0 %   ;
;     [C] Estimate of ALMs unavailable [=a+b+c+d]             ; 1,873 / 32,070      ; 6 %   ;
;         [a] Due to location constrained logic               ; 0                   ;       ;
;         [b] Due to LAB-wide signal conflicts                ; 442                 ;       ;
;         [c] Due to LAB input limits                         ; 1,431               ;       ;
;         [d] Due to virtual I/Os                             ; 0                   ;       ;

From the fitter report snippet above, I would assume that the 5,521 ALMs used only as registers [A.c] would be candidates for packing with the 11,471 ALMs used only as LUT logic [A.b].

Is there a reason that the fitter fails to pack any ALMs in no-fit cases?

Is there a Quartus option that would force the fitter to attempt to pack ALMs?