Mutate Top N Rows Without Throwing Away the Other Rows

We can group by date then sort the return variable get the last 3 enteries (top 3) and return return/n or else 0.

library(dplyr)
n <- 3
df %>%
  group_by(date) %>%
  mutate(w = ifelse(return %in% tail(sort(return), n), return/n, 0))


# date       industry return     w
#   <date>     <chr>     <dbl> <dbl>
# 1 2015-05-01 Hlth       4.89  1.63
# 2 2015-05-01 Txtls      4.37  1.46
# 3 2015-05-01 BusEq      4.02  1.34
# 4 2015-05-01 Fin        2.99  0   
# 5 2015-05-01 ElcEq      2.91  0   
# 6 2015-05-01 Food       2.03  0   
# 7 2015-05-01 Beer       2     0   
#....

The base R equivalent of the same logic using ave

ave(df$return, df$date, FUN = function(x) ifelse(x %in% tail(sort(x), n), x/n, 0))

EDIT

As mentioned in comments, in case of ties OP wants to return (1/n)/2 or divide by number of ties we have.

For this I have created a new easier dataframe which makes it easy to understand what is going on.

df <- data.frame(A1 = rep(c("A", "B"),c(5, 4)), A2 = 1:9)
df$A2[2] <- 3

If we use the current code it gives

df %>%
   group_by(A1) %>%
   mutate(w = ifelse(A2 %in% tail(sort(A2), n), A2/n, 0))

  # A tibble: 9 x 3
  # Groups:   A1 [2]
#  A1       A2     w
# <fct> <int> <dbl>
#1 A         1  0   
#2 A         3  1   
#3 A         3  1   
#4 A         4  1.33
#5 A         5  1.67
#6 B         6  0   
#7 B         7  2.33
#8 B         8  2.67
#9 B         9  3   

which is not what we want. To avoid that, we can group by A2 again and for only those rows where w!=0 we divide it by number of occurrences of A2.

df %>%
  group_by(A1) %>%
  mutate(w = ifelse(A2 %in% tail(sort(A2), n), A2/n, 0)) %>%
  group_by(A2) %>%
  mutate(w1 = ifelse(w != 0, w/n(), w)) %>%
  ungroup()

# A1       A2     w    w1
#  <fct> <dbl> <dbl> <dbl>
#1 A         1  0     0   
#2 A         3  1     0.5 
#3 A         3  1     0.5 
#4 A         4  1.33  1.33
#5 A         5  1.67  1.67
#6 B         6  0     0   
#7 B         7  2.33  2.33
#8 B         8  2.67  2.67
#9 B         9  3     3   

Another EDIT

Turns out we just want to divide w only for the last group present. Moreover, the sum of all the w in each group should sum up to 1. For the updated dataset we can do

n <- 3

temp_df <- df %>%
            group_by(A1) %>%
            top_n(n, A2)


 temp_df %>%
     arrange(A1, A2) %>%
     mutate(w = ifelse(A2 == A2[1], 
    (1 - (1/n * sum(A2 != A2[1])))/sum(A2 == A2[1]), 1/n)) %>%
     bind_rows(anti_join(df, temp_df) %>%
                      mutate(w = 0)
    ) %>%
     arrange(A1, A2)


# A1       A2     w
#  <fct> <dbl> <dbl>
#1 A         1 0    
#2 A         3 0.167
#3 A         3 0.167
#4 A         4 0.333
#5 A         5 0.333
#6 B         6 0    
#7 B         7 0.333
#8 B         8 0.333
#9 B         8 0.333

Let's try another variation where we keep all the values of the group same.

df1 = df
df1$A2[6:9] <- 10


 temp_df <- df1 %>%
             group_by(A1) %>%
             top_n(n, A2)


  temp_df %>%
       arrange(A1, A2) %>%
       mutate(w = ifelse(A2 == A2[1], 
      (1 - (1/n * sum(A2 != A2[1])))/sum(A2 == A2[1]), 1/n)) %>%
       bind_rows(anti_join(df1, temp_df) %>%
                      mutate(w = 0)
       ) %>%
       arrange(A1, A2)


#  A1       A2     w
#  <fct> <dbl> <dbl>
#1 A         1 0    
#2 A         3 0.167
#3 A         3 0.167
#4 A         4 0.333
#5 A         5 0.333
#6 B        10 0.25 
#7 B        10 0.25 
#8 B        10 0.25 
#9 B        10 0.25 

The logic is we select the top 3 A2 values along with their groups using top_n. Using anti_join we get all the rows which are not in top 3 and assign a fixed weight w to them as 0. For the rows which are included in top 3 we get the last group rows and assign them the weight which is remaining after assigning the weights to non-last groups.