Na minha versão da multicomp no Ubuntu 16.04 com R não deu warnings.

> modeloPARSUBlme <- lme(PRODLEIT ~ BLOCO + PASTO * MINERAL,
+                        random = ~1 | PARCELA,
+                        data = PARCSUB)
>
> compara <- glht(modeloPARSUBlme, linfct = mcp(PASTO = "Tukey"))
Warning message:
In mcp2matrix(model, linfct = linfct) :
  covariate interactions found -- default contrast might be inappropriate
> summary(compara)

     Simultaneous Tests for General Linear Hypotheses

Multiple Comparisons of Means: Tukey Contrasts


Fit: lme.formula(fixed = PRODLEIT ~ BLOCO + PASTO * MINERAL, data = PARCSUB,
    random = ~1 | PARCELA)

Linear Hypotheses:
           Estimate Std. Error z value Pr(>|z|)  
2 - 1 == 0   0.3333     1.4829   0.225  0.99600  
3 - 1 == 0   1.0000     1.4829   0.674  0.90684  
4 - 1 == 0   4.6667     1.4829   3.147  0.00899 **
3 - 2 == 0   0.6667     1.4829   0.450  0.96973  
4 - 2 == 0   4.3333     1.4829   2.922  0.01821 *
4 - 3 == 0   3.6667     1.4829   2.473  0.06416 .
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Adjusted p values reported -- single-step method)

>
> packageVersion("multcomp")
[1] ‘1.4.6’
> R.version.string
[1] "R version 3.3.1 (2016-06-21)"
>
> browseURL("https://cran.r-project.org/web/packages/multcomp/index.html")

À disposição.